zachary

same. it did not show the belief of prediction as the TabNine ads. I am not sure whether it is activated.

> device = 'cuda' model_name = "baichuan-7b" adapter_name = "weights/baichuan-7B/checkpoint-500" model = AutoModelForCausalLM.from_pretrained( model_name, trust_remote_code=True, low_cpu_mem_usage=True, torch_dtype=torch.float16, device_map='auto' ) tokenizer = AutoTokenizer.from_pretrained( model_name, trust_remote_code=True ) model = PeftModel.from_pretrained(model, adapter_name) model.eval()...

对的，我手动把外面的adapter_config.json拷贝进去了。我发现使用checkpoint里的adapter预测结果，跟使用最终的adapter预测结果差别很大，后者可以遵从指定微调的格式，而checkpoint里的adapter效果很差，就跟没加载adapter一样。 比如，后者生成的结果是：“否，需勾选【通过】” 用checkpoint生成的结果是空。调大了`max_new_tokens`也不行呢

> 119行：model.enable_input_require_grads()

> @ZacharyWaseda 解决了没 解决了。那俩文件换成最新的就行了